METHOD FOR TRAINING A CLASSIFIER DESIGNED FOR DETERMINING THE CATEGORY OF A DOCUMENT Russian patent published in 2018 - IPC G06K9/00 G06F17/30 

FIELD: information technology.

SUBSTANCE: invention relates to prevention of information leaks, in particular to the prevention of leaks of electronic copies of personal and confidential documents. In the method of training a classifier intended for determining the category of a document, documents that belong to the category are received. For each document received, the objects contained in it are defined, which are graphic elements. For each document received, a set of characteristics consisting of certain objects is formed. In this case, the mentioned characteristics are the characteristics characterizing the presence of an object, the location of an object, the number of objects, the location of one object relatively to another object, the size of an object, the angle of object's inclination. Construction of a classifier based on the values of the generated characteristics for the received documents is created.

EFFECT: technical result is higher quality of a category definition of a document by a classifier.

13 cl, 8 dwg